Offshore wind operation and maintenance is modelled by two families of tools that do not meet. Discrete-event simulation carries the mechanisms that drive offshore downtime but costs many runs per converged answer, whereas the published analytical models answer for a fraction of the computational cost yet are exponential, steady-state, and window-based, and the effects of these assumptions have not been benchmarked against a modern simulator.

This thesis develops a hierarchy of analytical availability models that occupies that gap, establishing what each member can be trusted to estimate, at what accuracy, and at what cost. The family pairs steady-state and time-marching solvers under Continuous Time Markov Chain and a Semi-Markov Process frameworks, plus two simulator twins, all consuming one shared input pipeline. The reference case is a bottom-fixed 12MW turbine with 22 corrective failure modes over twenty years of hourly ERA5 weather, and the discrete-event reference, NREL’s WOMBAT (0.13.3), is audited, minimally patched, and characterised through a pre-registered mechanism-accretion ladder.

Under aligned conditions the steady-state semi-Markov model reproduces the reference raw, at z = +0.20 over 200 seeds and z = −1.02 against the thousand-seed population, certifying 97.9903% time-based availability, 97.8243% production-based availability, and an annual direct cost of 576,393.60USD for the selected case study turbine. The memoryless simplification is evaluated and proves adequate for mean-determined steady-state quantities, although it cannot express production-based availability accurately, overstates the cold-start boundary credit by 46%, and is blind to the replacement waves that make a twenty-year life
at β = 2 one long transient, missing 1.47 percentage points of year-one availability. Process interruptibility treatment is also evaluated, with results indicating a second-order convention worth 0.167 percentage points for the short repairs, and the enabling assumption for heavy replacement modelling, whose contiguous execution diverges at the reference site.

Finally, the 13 ms solve enables rapid scenario screening and sensitivity analysis, being used to propagate forty thousand joint uncertainty samples, price vessel capability, lifting envelopes, mobilisation lead time, and component reliability with break-evens attached, therefore turning the validated hierarchy into a decision instrument at a cost measured in seconds. ... Read more

The rapid technological progression of wind turbines not only imposes design challenges but also necessitates continuous advancements in modeling approaches, particularly holistic methods capable of accurately capturing turbine coupling effects. A well-established modeling tool in the wind energy sector is the aero-servo-elastic code HAWC2. Although it is widely used to predict the overall turbine response across a broad range of operating conditions, the drivetrain is typically represented as a single beam.
In the presented work, a more realistic and comprehensive drive train model was developed in HAWC2 to account for additional flexibility effects associated with increasing turbine dimensions and enable the estimation of the main bearing reaction loads. First, a detailed literature review was conducted to analyze current trends regarding drivetrain technology, essential components, and modeling approaches of modern wind turbine drivetrains.
The DTU 10MWRWT and a high-fidelity SIMPACK drivetrain model, developed by researchers from NTNU, were identified as references for the implementation. Additionally, the theoretical foundations of Timoshenko beam elements and multibody formulation used in HAWC2 were studied, as both are essential for constructing a turbine model in HAWC2. In the implementation phase, the overall drivetrain structure and fundamental drivetrain properties, such as the stiffness properties of the main bearings, were extracted from the SIMPACK model and incorporated into HAWC2 through the introduction of additional beam elements. To further increase the fidelity, the first torsional eigenfrequency of the derived structure was tuned to match the dynamics of the reference models. As a final step, the simple HAWC2 drivetrain representation was adjusted to reflect the mass distribution of the developed structure, enabling a direct comparison between the two.
The developed model showed strong performance in predicting main bearing loads under steady conditions, thereby achieving a clear improvement in fidelity compared with the original setup. In contrast, reduced agreement was observed for the torque arms response and under turbulent simulations. The study concludes with a critical evaluation of the model, addressing its limitations and outlining potential directions for further accuracy improvements.
... Read more

Amid the global energy transition, offshore wind energy has been positioned as a pivotal technology,
utilizing both bottom-fixed (BOWT) and floating offshore wind turbine (FOWT) configurations. Offshore wind turbines face high operation and maintenance (O&M) costs, representing up to 35% of the levelized cost of energy (LCOE), with the drivetrain and especially the main bearing (MB) being among the most failure-prone components, leading to long downtime and reduced energy production. In floating wind turbines, main bearings experience additional challenges due to combined wind and wave loads and the platform motions.

This research investigates the differences in loading and fatigue damage of the upwind main bearing (MB1) in a monopile BOWT and semi-submersible FOWT, both based on the DTU 10 MW reference wind turbine (RWT). Using OpenFAST simulations under Design Load Case (DLC) 1.2, according to Standard IEC 61400-1: 2019, this study calculates the MB1 loads for both wind turbine configurations, at a North Sea location east of Scotland, with a water depth of 60 m. Simulations are conducted across the full operating wind speed range (4.5 - 24.5 m/s), and loads are calculated through an analytical formulation. The axial and radial loads of MB1 are post-processed according to Standard ISO 281:2007, to compute the dynamic equivalent load, which is used to evaluate the fatigue damage of the main bearings. The Load Duration Distribution (LDD) method, which defines the load cycles, and the site-specific Weibull distribution are used to estimate the hourly fatigue damage of MB1, which is then extrapolated to a 20-year design lifetime, yielding the time-dependent Remaining Useful Life (RUL).

A comparative analysis of MB1 load means and standard deviations reveals that, for the examined
drivetrain configuration, radial loads dominate the main bearing loading and fatigue contribution in both configurations. Specifically, for the FOWT, they represent 74 to 90% of the dynamic equivalent load and 39 to 72% of the fatigue damage across all operating wind speeds. For the BOWT configuration, the corresponding ranges are 75 to 90% and 42 to 72%, respectively. MB1 of the FOWT experiences higher axial and radial loads, mainly due to platform surge and pitch motions, which increase rotor thrust and amplify the axial loading, as well as platform sway, roll and yaw motions that mainly amplify the radial loads. The largest load differences between FOWT and BOWT are encountered at around the rated wind speed of 11.4 m/s, and they peak at 13.5 m/s, where the FOWT axial load is 13.17% higher than in the BOWT.

Over the 20-year lifetime, MB1 in the FOWT accumulates 5.03% more fatigue damage, failing at 11.98 years, compared to 12.58 years for the BOWT. BOWT MB1 exhibits higher fatigue damage in the below-rated wind speed region, especially for wind speeds below 8.5 m/s. For aligned wind-wave cases, the reduction in the MB1 damage for both FOWTs and BOWTs is only 0.6% compared to realistic (misaligned) cases, indicating that, the wind-wave misalignment for wind speeds in the above-rated region, has a slight influence on the cumulative damage. A parametric sensitivity analysis shows that turbulence intensity is the most influential environmental factor: a change from Class A to C results in up to 10% reduction in MB1 damage. Among system parameters, platform mass significantly affects MB1 fatigue in FOWTs, with a 20% increase in mass reducing damage by 5.6%, while a 20% decrease increases it by 4.4%.

Overall, the findings of this thesis prove that MB loads in floating wind turbines are subjected to greater complexity and amplification, due to platform motions and aerodynamic, hydrodynamic, structural and control coupling. This results in increased fatigue damage and indicates the importance of advanced fatigue-aware, system-integrated control strategies that consider the platform dynamics to ensure main bearing reliability in future floating wind developments. The study concludes with recommendations for future work, including the validation of the results with field data, the exploration of alternative drivetrain layouts and the implementation of advanced control schemes to mitigate the impact of the platform motion. ... Read more

Wind energy plays a pivotal role in the energy transition, however, a major obstacle for offshore wind farms remains the high Operations and Maintenance (O&M) costs, which can be up to 35% of the levelized cost of energy (LCOE). Traditional maintenance strategies often result in excessive downtime, increased costs, or premature component replacements. To counter these problems, condition-based maintenance (CBM) uses real-time data to estimate machine health and guide maintenance decisions. Although most CBM studies address either health assessment or maintenance optimisation in isolation, this thesis proposes an end-to-end CBM framework that derives health indicators (HIs) and integrates them into a multiple-threshold CBM framework.
The framework is validated on rotor bearing failures from the CARE data set, which supplies real-world Supervisory Control and Data Acquisition (SCADA) measurements. Rotor bearings, despite their relatively low failure frequencies, cause high downtime and exhibit a prolonged degradation pattern. HIs are extracted with a long- and short-term memory (LSTM) autoencoder (AE) trained solely on healthy turbines to learn normal behaviour. On avarage, the resulting HIs identify anomalies 195 days before failure.
These HIs are fed into a CBM strategy that prioritises timely, minor interventions over costly lastminute replacements. Compared to a purely corrective replacement policy, the strategy reduces annual maintenance expenditure on rotor bearings by an average of 62.5%, mainly by prolonging the useful life of the components and reducing downtime. This thesis therefore demonstrates the value of rotor-bearing HIs, derived from widely available SCADA data, for wind turbine maintenance.
... Read more

Offshore wind farms face significant operational and financial challenges due to weather-related uncertainties that disrupt short-term maintenance planning. Operational decisions are critical to maintaining turbine performance and ensuring technician safety. When such decisions fail to effectively account for short-term risks like adverse weather and human limitations, they can result in failed maintenance attempts, prolonged downtime, increased CO2 emissions, and substantial financial losses.

This thesis develops a probabilistic decision-support model designed to improve the reliability, cost efficiency, safety, and sustainability of offshore wind maintenance operations by incorporating short-term weather forecasts, vessel operability constraints, and crew safety—specifically, the risks of seasickness during transit to the offshore wind farm (OWF) and unsafe technician transfer from vessel to turbine platform.

The model was evaluated through a 10-year case study simulating 19,090 minor repair tasks executed by a Crew Transfer Vessel (CTV), drawn from 75 maintenance schedules across 10 turbines, for a potential OWF site located approximately 40 km offshore from Cabo Silleiro, Spain, in the Atlantic Ocean. Hindcast weather data (2001–2010) for the offshore location were used to compare two maintenance scheduling strategies: a standard industry approach using deterministic vessel operability thresholds and fixed crew size (Case 1), and the proposed probabilistic model integrating transit and transfer uncertainties, a cost-loss decision framework and dynamic crew size optimization (Case 2).

Mission feasibility was determined using the combined probability of mission success, calculated as the product of (i) transfer success probability, based on wave height, wind speed, and wave period, and (ii) transit success probability, modeled using a Binomial distribution, to estimate the likelihood that enough technicians remain healthy (not seasick) upon arrival at the turbine. This probability was derived using Motion Sickness Incidence (MSI) empirical values and sea state conditions. A mission was attempted only if this combined probability exceeded the cost–loss ratio, which compares the cost of a mission attempt with the expected loss from failure. The two strategies were evaluated based on key performance metrics, such as combined probability of mission success, expected operational costs, and environmental impact (CO2 emissions). A sensitivity analysis further confirmed the model’s robustness
across different weather conditions by testing three scenarios: Scenario 1 (20% harsher than hindcast weather), Baseline scenario (hindcast weather), and Scenario 2 (20% more favorable weather).

Results showed that Case 2 improved the average combined probability of mission success to 68.7%, compared to 43.3% in Case 1 (a 58% increase), and reduced expected financial losses by approximately $540,000 over the planning horizon. It also reduced CO2 emissions from re-attempted trips by up to 45.8%. Although Case 2 introduced an average delay of 35 days in baseline conditions by avoiding risky missions, it consistently maintained a success rate above 67% even in the worst-case scenario, demonstrating reliable and robust performance across diverse weather conditions.

The study acknowledges several limitations. Due to confidentiality constraints, real-world maintenance records were not available for validation. Seasickness was estimated using empirical MSI values under simplified assumptions. The model also assumes a fixed start time for daily shift, a single task per day using one CTV, and does not currently account for task prioritization. Despite these constraints, the framework is adaptable and suitable for real-world implementation.

Future research could focus on validating the model using industry maintenance data, developing a new metric for seasickness estimation, and extending the scheduling logic to support multiple vessels and dynamic shift planning. Broader applications could include SOV-based operations, major repairs, and prioritization of critical tasks.

In conclusion, this thesis presents a practical, flexible, and sustainability-oriented framework for offshore wind maintenance planning. By enabling smarter, risk-informed decisions, the model supports reduced unnecessary vessel trips, lower emissions, and more efficient use of resources. ... Read more

As the world shifts toward sustainable energy, offshore wind power is becoming increasingly important because of its ability to capture stronger and more consistent winds. However, offshore wind turbines operate in harsh environmental conditions, exposing wind turbine blades to loads. These conditions accelerate blade degradation through mechanisms such as leading-edge erosion, which disrupt airflow, increase drag, and decrease lift, ultimately reducing the aerodynamic efficiency and energy output of the turbine. Over time, this degradation alters the aerodynamic characteristics of the turbine, posing
significant challenges to control strategies.

This work addresses this challenge by proposing a physics-informed, learning-based framework for modeling blade degradation in wind turbines. Unlike earlier studies that model degradation as a scalar reduction in performance (typically as a uniform downscaling of the power coefficient CP ), the method developed here decouples the impact on lift and drag coefficients. This approach enables the reconstruction of complex, shape-altering changes in the CP (λ) curve. The result is a generalized degradation framework, parameterized through sensitivity coefficients k1 and k2, that represents the degradation
trajectory in a degradation severity space. These sensitivity coefficients are not static but evolve over time, enabling a dynamic and interpretable representation of aerodynamic degradation.

The framework is implemented into the Wind Speed Estimation - Tip-Speed Ratio (WSE-TSR) control scheme, a closed-loop scheme that is capable of learning. Within this control scheme, the degraded
turbine dynamics are compared with an internal reference model to derive estimation errors, which in turn are used to iteratively update the internal degradation parameters. The result is a correcting controller that learns degradation severity and tracks the evolution of the CP (λ) curve over time.

Initial simulation results, based on modeled degradation paths and a realistic leading-edge erosion case,
revealed limitations in the learning process caused by systematic errors in the control scheme. These errors led to discrepancies between the estimated and actual degraded aerodynamic profiles. However, after identifying and analytically correcting for this model error, the learning algorithm demonstrated significantly improved performance. It was then able to reconstruct accurate degraded CP (λ) curves for moderate and severe degradation scenarios. A discussion is provided on the origin of the learning
inconsistencies and how they could be further investigated. The framework underscores the potential of the framework to support integration into digital twin systems for adaptive control in a degraded state.

By combining physics-based simulation, parameter learning, and a control scheme, this thesis presents a framework that enhances our ability to monitor and interpret degradation in offshore wind turbines. In particular, it provides physics-based insight into how degradation mechanisms, such as leading-edge
erosion, affect aerodynamic performance through changes in lift and drag. This approach contributes to the development of more accurate and adaptive digital twin models that can account for suboptimal aerodynamic performance due to blade degradation. ... Read more

Offshore wind turbines (OWTs) play a critical role in renewable energy, requiring efficient methods to predict fatigue loads on their support structures under harsh environmental conditions. Traditional fatigue assessment methods are effective but costly and impractical at scale. Surrogate models (SMs) offer a cost-effective alternative, but site-specific SMs often fail to generalize across turbines in varying environments and locations. This study explores whether platform-generic SMs trained on a Simulation Database can serve as reliable replacements for site-specific SMs trained on Supervisory Control and Data Acquisition (SCADA) and Structural Health Monitoring (SHM) data for predicting fatigue loads in OWT support structures. The SMs, developed using both deterministic and Bayesian neural networks (NNs), were evaluated in three progressively complex model setups featuring SCADA signals, nacelle accelerations, and tower top strain gauges. Results showed that site-specific SMs generally achieved lower mean average percentage error (MAPE), with deterministic NNs at 47.1m lowest astronomical tide (LAT) in the fore-aft (FA) direction reducing errors from 16.8% to 7.2% and in the side-side (SS) direction from 27.0% to 4.5% as additional signals were introduced. Platform-generic SMs exhibited higher MAPEs due to the broader scope of the model and slight mismatches between simulated and real turbine dynamics, especially in the FA direction. Nonetheless, in the SS direction at 47.1m LAT, platform-generic SMs showed promising performance, achieving errors as low as 7.5% in one configuration. Although Bayesian NNs did not consistently lower errors compared with deterministic approaches, they provided valuable insights into how far test data deviated from the training distribution, helping identify the potential limits of the model. This capability has significant practical implications, as it can potentially serve as a tool for detecting sensor malfunctions or identifying irregular data, ensuring more reliable predictions in real-world applications. Overall, platform-generic SMs still require refinement before they can fully replace site-specific SMs, and Bayesian NNs should not replace deterministic NNs but rather serve as a valuable supplement. ... Read more

A key hurdle to meet the ambitious renewable energy targets is the high levelized cost of offshore wind energy. With Operations and Maintenance (O&M) actions making up more than a quarter of the total costs, substantial research efforts have been undertaken to optimize these actions. In line with this overarching goal, this project focuses on O&M actions, specifically optimizing time-based maintenance strategies to minimize costs. Literature reveals a variety of methods developed, such as Remaining Useful Life (RUL) models, degradation models, reliability models and maintenance models to simulate Wind Turbine (WT) lifetimes, O&M actions and calculate costs. A majority of these studies only look at
single failure modes (FM) or multiple independent FMs. In the field of wind energy, there is a significant lack of research on failure interactions and degradation dependence between different components or FMs. This project aims to build a model that accounts for dependencies in degradation between different FMs. This model is further combined with a maintenance model to simulate maintenance actions and estimate corresponding O&M costs.

The proposed degradation-based maintenance model builds on a simplified in-house model developed by the Wind Energy group at TNO. A Markov model, built in Python, is used to represent the degradation process. This is further combined with copula functions to represent degradation dependencies, using available copula packages in Python. Inspection and maintenance functions are then integrated into the degradation model to simulate O&M actions within the WT lifetime. Each action incurs either one or a combination of costs due to inspection, maintenance or operation, from WT downtime. An extensive number of time-based strategies are modelled, with varying inspection intervals for each. Finally, the accumulated lifetime costs are calculated and compared for the various maintenance strategies. The inspection interval with the minimum lifetime costs is selected as the optimum maintenance
strategy. In order to verify the validity of the proposed model, the results of an existing research paper on degradation modelling for WT blades are first replicated.

After successful verification, case studies are carried out to apply the model to the degradation processes of offshore wind turbine (OWT) hydraulic pitch systems. This is an extremely critical system, crucial for both WT operation and safety with the highest failure rates among all WT systems. The case studies selected in this work cover two pitch FMs: hydraulic fluid leakage and hydraulic valve wear. The model is applied to the case of single FMs and the case of multiple FMs, both independent and dependent. For both the single FM models, different optimum inspection intervals and lifetime costs are obtained. A higher failure rate for one of the FMs is the basis for the increased optimum lifetime cost and the decreased optimum inspection interval. Comparing the direct summation of the results of two single FM models with the multiple FM independent model outputs dissimilar results. The independent model leads to a reduced inspection interval and slightly higher lifetime costs. For the dependent
case, a 39% reduction in the lifetime costs and an increase in the optimal inspection interval is
observed compared to the independent case. This results in cheaper and less frequent O&M actions for the dependent case.

This work shows that, compared to the other case studies, modelling dependency between FMs allows for improved O&M actions, resulting in lower costs. This methodology has the potential to be used in future multi-component or multi-FM studies. The inclusion of degradation dependencies between components would better model the inter-dependencies between FMs and hence allow for improved optimization of O&M strategies. ... Read more

The main bearing is a critical wind turbine drivetrain component, and its failure can cause the turbine shutdown and expensive repair. As the main bearing degrades, its temperature increases, indicating health deterioration of the component. Various methods proposed in the literature employ physics-based normal behavior models (NBMs) using Standard Supervisory Control And Data Acquisition (SCADA) data as input to energy conservation-based equations to model the main bearing temperature. These methods analyze the difference between the measured and the modeled temperature, referred to as residual. These residuals are used as health indicators (HIs) for assessing the condition of the component.

Physics-based NBMs utilizing SCADA data have been successfully used for fault detection of the main bearing. However, the application of these methods for degradation trend monitoring have not yet received attention. The primary reason for the premature failure of wind turbine components is attributed to the variability of the wind conditions. However, current NBM methods are based solely on the mean value records and do not consider the variations within the 10-minute time frame. Furthermore, seasonal fluctuations in operating conditions can adversely affect the obtained degradation trend.

The main objective of this thesis is to improve physics-based NBM employing SCADA data to monitor the degradation trend of the main bearing. The proposed approach uses a physics-based NBM available in the literature as the baseline. It aims to increase the monotonicity and reduce the dispersion of the developed HI to enable accurate degradation trend monitoring. To achieve this objective, the proposed method takes into account seasonal variations and variability of operating conditions within the 10-minute SCADA time frame when modeling the main bearing temperature. To mitigate the impact of seasonal changes on the HI, the proposed method develops multiple physics-based NBMs corresponding to monthly time windows. To take into account the variability of the operating conditions, the main bearing temperature is modeled by performing a Monte Carlo simulation using the SCADA data mean and standard deviation values. In this case, the HI is defined by the data density within a threshold region. Two case studies are conducted to demonstrate the advantages of the proposed method compared to the baseline approach. The results show that with the proposed approach, the seasonality effects are reduced by more than 50%, as measured through cross-correlation metric with the ambient temperature, the HI monotonicity increases by 260% as measured by the Mann-Kendall τ monotonicity metric, and the dispersion reduces by 30% and 35%, as evaluated by the Mean Square Error and a noise metric obtained using the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise approach. ... Read more

Accurate fault detection in wind turbines is a key factor for improving their reliability and reducing maintenance cost. This report investigates probabilistic target variable modeling using a conditional Generative Adversarial Network (cGAN) in a Normal Behavior Modeling framework for fault detection on wind turbine drivetrains. The dataset used for the investigation is the open-source Supervisory Control and Data Acquisition (SCADA) dataset provided by Energias de Portugal (EDP). A cGAN and a traditional Recurrent Neural Network (the base model) are developed and trained to model the generator bearing temperature. The performance of the cGAN is compared to that of the base model, to highlight differences associated with the modeling approach. The cGAN’s output is a conditional distribution of the target temperature, given the operational state of the turbine. From this distribution, two methods of deriving a Health Indicator are investigated: (1) By reducing the distribution to a point estimate, and (2) by taking into account the whole distribution. From the base model, a third Health Indicator is derived.
For all three Health Indicators, anomalies are detected by performing a non-parametric hypothesis test designed to detect deviations from the empirical healthy data distribution. The findings indicate an enhanced temperature modeling performance of the cGAN on train, validation and test data. Especially the robustness appears to be significantly improved. While interpreting the outcome of applying the Health Indicators for fault detection remains challenging, the results suggest that the cGAN’s probabilistic modeling improves fault detection by reducing the number of false responses. ... Read more

Rotor imbalances—such as mass imbalance, pitch misalignment, and yaw misalignment—are critical faults in wind turbine systems. These imbalances cause uneven load distribution on components, leading to excessive wear, failures, increased operational costs due to unplanned downtime, and reduced energy output. Despite advancements in monitoring technologies, current maintenance strategies in wind turbines still rely on time-based manual inspections, as they lack reliable automated detection systems. This thesis addresses the need for a more efficient fault detection framework by integrating already available drivetrain Condition Monitoring System (CMS) vibration signals— commonly used to detect drivetrain component failures, like in gears and bearings— with traditional SCADA data. The aim is to extract signal features that capture the system's dynamic behavior and effectively detect and diagnose rotor imbalances. Notably, this approach overcomes the limitation of current systems not having direct measurements from the blades by leveraging operational data already collected from wind turbines.

Building on prior research, the proposed approach combines frequency and time-domain analyses and focuses on two key data sources: drivetrain vibration measurements and rotor speed data from the SCADA system. A decoupled simulation framework integrates aeroelastic simulations from OpenFAST with a multi-body drivetrain model in SIMPACK, specifically for the 10 MW DTU reference wind turbine. The results show that drivetrain velocity signals, particularly in the side-to-side direction, are highly sensitive to rotor imbalances, enabling accurate trend analysis. Features such as peak amplitudes at 1P and 3P frequencies form the basis of the fault detection and diagnosis criteria proposed in this thesis. By using the median values of their distributions, imbalances can be effectively detected and diagnosed. This approach also supports the implementation of a decision tree framework for real-time fault classification across various operating conditions.

The methodology was tested under both above and below-rated wind speeds, first in steady-state conditions and then in turbulent inflow scenarios. Additionally, health state indicators are proposed to recognize fault severity levels by clustering median value features within predefined ranges for low, medium, and high severity. This comprehensive monitoring approach effectively tracks fault progression across the imbalance scenarios under study. As a result, the proposed method lays the foundation for a future data-driven system that can reduce reliance on manual inspections and provide a scalable solution for predictive maintenance in wind turbine operations. ... Read more

The electricity generation from wind has seen significant growth over the past two decades with a compound annual growth rate of over 21%, and this trend is projected to continue, aligning with the goals of the European Green Deal. In the context of offshore wind turbines, operational and maintenance costs can account for up to 30% of the total costs throughout the project's lifecycle. This highlights the need for the development of efficient condition monitoring methods aimed at mitigating maintenance expenses.
The gearbox is one of the most failure-prone components in wind turbines, leading to extended downtimes and substantial financial outlays. When the components of the gearbox degrade, the heat generated within the gearbox increases. This increase leads to higher oil temperatures, making the temperature signal a suitable indicator of the gearbox health condition. The main objective of this thesis is to design a physics-based normal behaviour model (NBM) for the estimation of the wind turbine gearbox oil temperature and to use field data to validate its effectiveness. The energy conservation principle is applied to the gearbox to formulate an equation for the calculation of the gearbox oil temperature. To account for some unidentified design specifications of the gearbox, the equation is fitted to available historical data from the Supervisory Control and Data Acquisition (SCADA) system to determine the unknown parameters. The model uses as input an array of operational measured signals available in the SCADA system, including rotor speed, power output, nacelle temperature and inlet oil temperature, which is the temperature of the oil after running through the cooling system.
A case study demonstrates the model's effectiveness in accurately predicting the gearbox oil temperature with a mean absolute percentage error of 0.75%. In order to investigate how well the physical characteristics of the gearbox are represented by the model, the coefficients of the energy balance equation derived from fitting it to the field SCADA data are compared to the parameters calculated using known characteristics of a reference gearbox. The comparison results indicate that the model's parameters are generally within the same order of magnitude as the reference values. To provide a benchmark, the performance of the physics-based model is also compared to that of two data-driven models using artificial neural networks (ANNs) for temperature prediction proposed in the literature. The results show that the proposed physics-based model outperforms both ANN models, achieving 14% and 59% lower root mean squared error (RMSE) and a reduction in time required for training of 70% and 95% when compared to the two ANN NBMs respectively.
... Read more

With growing wind energy capacity, especially offshore, reliability of wind turbines (WT) becomes a relevant concern. Poor reliability directly affects their cost effectiveness due to increased operation and maintenance (O&M) costs and reduced availability to generate power because of downtime. This certainly encourages WT operators to employ advanced O&M methodologies and focus on the critical components to reduce failure rate, time to repair and maximizing WT performance. Condition monitoring (CM) of wind turbines for the purpose of prognostics and health management of critical equipments can improve the reliability and reduce maintenance costs by identifying failures at the earliest possible stage and by eliminating unnecessary scheduled maintenance. In contrast to the expensive purpose-built condition monitoring systems, a SCADA (Supervisory Control and Data Acquisition System) data-based condition monitoring system uses data already collected at the wind turbine controller and provides a cost-effective way to monitor wind turbines.

This research focusses on developing a prognostics framework for WT gearboxes, which are one of the costliest subsystems to maintain during a turbine’s life. The framework follows a data-driven approach and combines two machine learning algorithms – Artificial Neural Network and Support Vector Machine to capture anomalous operations of the WT gearbox.  A real-time monitoring scheme is developed to track the degradation and set a maintenance alarm as the first evident signature of failure is identified.  The framework was implemented using high-frequency SCADA data and was able to detect gearbox failure, a month in advance, providing enough lead time to plan and perform required maintenance activities. Additionally, a sensitivity study is conducted to determine an optimal sampling frequency of SCADA data which can be used for CM purposes as the current industry practice of storing it as 10 min averages leads to a loss of information about the condition of a WT component. 

The results show that the feed­forward ANN can efficiently learn the complex mapping between the input and output features. To analyse the error between ANN predictions and the in-field measurements, four residual error features ­ maximum error, minimum error, root mean squared error and error distribution are used as inputs for the OC­SVM model to understand the complex boundary between normal and anomalous operation. The percentage of anomalies computed for each week of operation, 4 months before failure, show an increasing trend as the turbine approaches failure. To determine a threshold for maintenance alert, a real­time monitoring scheme based on linear regression and bootstrapped confidence intervals is developed to track the progression of anomalies and alarm a maintenance alert as the first indication of incipient fault becomes evident. The scheme alarms for maintenance a month before the actual failure, providing enough lead time to plan and maintain the gearbox.

A sensitivity study is carried out for a range of sampling periods ranging from 100 Hz to 10 min. The results demonstrate that high­frequency SCADA data is beneficial for condition monitoring of the gearbox, but only if the noise in the data can be excluded. On the other hand, despite the loss of information due to the averaging effect for large sampling periods, SCADA data aggregated over a 30 s period could be utilized to predict the gearbox failure a month in advance. Furthermore, the ANN model performance is found to be sensitive to the number of data samples available for training.

... Read more

With the rapid growth of offshore wind energy capacity, the operations and maintenance (O&M) challenges have increased significantly. Prognostics is the ability to know the condition of an equipment and to plan and perform maintenance prior to critical failure. The generator of the wind turbine is one of the most expensive components and therefore accurate prognostics of the generator can reduce costs of O&M. The existing solutions for prognostics rely on expensive, purpose built condition monitoring systems. This research presents a prognostic method to predict the remaining useful life (RUL) of generators, that uses no additional hardware beyond the standard SCADA systems installed in turbines. By applying machine learning techniques to detect anomalies in the data, the health of the generator is quantified into an Anomaly Operation Index (AOI). The degradation of the generator is then studied using a time series analysis method to estimate the generator’s RUL. The experimental study on real-world wind turbine data shows that this method predicts the RUL of the generator with good accuracies and provides sufficient lead times to the operators to schedule maintenance and repair. ... Read more

Increasing awareness about climate change and increasing interest in renewable energy is fueling the rise of wind energy. One of the main challenges currently faced by the wind energy industry is to improve the reliability and availability of wind turbine in order to keep the industry financially attractive. Optimising operations and maintenance (O\&M) strategy through the adoption of cost-effective and reliable detection and prognosis techniques is a clear target for competitive offshore wind development. If the health of components can be accurately monitored then faults can be detected in the early stages and an accurate maintenance plan can be made. This thesis will contribute to this domain of wind energy with original work. This thesis aims to devise multiple AI techniques capable of performing wind turbine fault detection. These techniques are validated and assessed using experimental data from the wind turbine drive train condition monitoring test rig developed at Durham University which focuses on the most critical wind turbine component, i.e. gearbox. Seeded-fault conditions on the gearbox have been induced/removed from the test rig drive train as required, enabling gearbox tooth damage fault to be implemented repeatedly on demand and under controlled stationary and variable driving conditions. The raw data from test rig is subjected to data pre-processing tasks such as data cleaning, feature selection and feature extraction. One baseline model (Decision tree), two tree-based models (Random forest and eXtreme Gradient Boosting) and one neural networks model (Multi-layer perceptron) are developed, tuned and compared to achieve the aim of this thesis. The performance of relatively unexplored (in CMS literature) tree-based models is compared to the performance of widely used neural networks model. Developed models are evaluated based on various performance criteria such as prediction of healthy stage, early stage fault detection, most severe stage fault detection, overall accuracy, overall precision, fault detection rate, false alarm rate, training time, testing time, complexity and fault detection (binary classifiers) performance. Three main models (Random forest, eXtreme Gradient Boosting and Multi-layer perceptron) outperformed the baseline model (Decision tree) in all important evaluation criteria. It is found that eXtreme Gradient Boosting (i.e, a tree-based model) produces the best overall results when compared to other developed models. Hence, it is recommended for the CMS industry to make use of the eXtreme Gradient Boosting machine learning model for the task of wind turbine gearbox fault detection. ... Read more